Conglomerate Discount: Why the Parts Can Exceed the Whole

A conglomerate discount is the distance between a group's divisions priced one at a time and the group's own traded value. For Sankalp Industrial Systems Limited, invented, that is Rs 24,03,30,00,000 of parts against Rs 22,40,00,00,000 traded: Rs 1,63,30,00,000, or 6.79 per cent measured off the parts. Four mechanisms produce it, and one chosen multiple produces two thirds of this particular figure.

Start in a household. The shape of this idea is much easier to feel there than in a set of accounts. Picture a home where three different things bring money in. One person draws a salary. Somebody else takes tuition classes in the evenings and is paid in cash. A back room is let out to a lodger. Three activities, three different rhythms, three different levels of risk, and every rupee any of them earns lands in the same bank account.

Now the awkward question. What are the evening tuition classes worth on their own? The question cannot be answered, and it is worth being precise about why. The classes have no accounts of their own; the takings are never separated from the salary once they hit the account. The classes run in a room the household already had, so they own no assets of their own. The classes cannot be bought without buying the household around them. And when the household decides to spend a year of tuition earnings on repairing the roof over the lodger's room, nobody outside is told and nobody outside could have stopped it.

Every one of those four obstacles reappears, in a more expensive form, when a listed group with several businesses inside it is priced by a market. The arithmetic below is exact and small; the four obstacles are the part worth carrying away.

What two numbers is a conglomerate discount the difference between?

Exactly two, and it is worth saying at the outset that they are built in completely different ways. The first is what the market pays for the whole company as it stands. Sankalp Industrial Systems Limited, invented throughout, is a listed manufacturer of industrial valves, precision castings and the aftermarket parts and service that go with them. Its traded enterprise value is Rs 22,40,00,00,000. The traded figure is observed. Somebody actually transacted at the price behind it.

The second number comes from an exercise called sum of the partsValuing each division on its own multiple and adding the results.. The group is taken apart on paper, a multiple is put on each division separately, the results are added, and an adjustment is made for the head office cost that belongs to no division. Building a sum of the parts is worked through in full under sum of the parts valuation, and it produces Rs 24,03,30,00,000. The parts figure is constructed. Nobody transacted at it, and four separate assumptions had to be chosen before it existed.

Subtracting one from the other gives Rs 1,63,30,00,000. Dividing that by the parts gives 6.79 per cent, the figure used throughout. A conglomerate discountThe gap between a group's traded value and its divisions valued separately. is that subtraction and nothing else, and it borrows every weakness of whichever of the two numbers was built rather than observed.

Which base is the percentage measured off, and does it matter?

The base matters more than most readers expect. The same rupee figure produces two different headline percentages depending on which figure it is divided by. Rs 1,63,30,00,000 over the parts of Rs 24,03,30,00,000 is 6.79 per cent. The identical Rs 1,63,30,00,000 over the traded Rs 22,40,00,00,000 is 7.29 per cent. Half a percentage point appears out of nothing but a choice of denominator.

The convention is to quote a discount off the higher of the two figures. Here the higher figure is the parts, and 6.79 per cent is therefore the number carried through everything below. But conventions are not laws, and somebody quoting 7.29 per cent has not made an error; they have used the other base. A discount quoted without its base named is a figure nobody can check, so the base belongs in the same sentence every single time. The same discipline applies to a premium, to a holding company discount and to every other ratio built by dividing a difference by one of the two things it came from.

Why is a difference between two numbers not by itself evidence?

Because of an asymmetry that gets lost the moment somebody writes the answer down as a percentage. The traded figure is a fact about a market. A reader may dislike it and may expect it to move, but nobody can argue with what it is. The parts figure is not a fact about anything. The parts figure is the output of a small model with four inputs: a multiple for each of three divisions, and a treatment for the head office cost.

Subtract a modelled number from an observed one and the result inherits the modelled number's uncertainty in full. If the parts figure could plausibly have been anywhere in a range, then so could the discount, and a discount that could plausibly have been zero is not a finding. The subtraction does not launder an assumption into an observation, however clean the resulting percentage looks.

None of that should collapse into cynicism. There are real, nameable, separable reasons why a group of businesses under one listing is priced below the same businesses priced apart, and those reasons exist whether or not any particular measurement of them is reliable. Understanding the mechanisms is the useful part. Quoting the measurement as though it settled something is the failure, and both of those are covered below.

Cause one: what happens when the market can only see a segment line?

Go back to the household for a second. The reason nobody can price the evening tuition classes is not that they are a bad business. The reason is that there is nothing to look at. There is no separate ledger, no separate bank statement, and the only evidence anybody outside has is that the household seems to be doing all right.

A division inside a listed group is in a better position than that, but not by as much as people assume. A listed group publishes segment disclosureThe part of a group's reporting that splits results by division. about its divisions: usually a revenue line and a result line for each division, sometimes a little more. A standalone listed company publishes a full set of statements about itself, audited on that company alone, with a balance sheet, a cash flow statement, borrowings in its own name and a price quoted on it every trading day.

Put those two objects side by side and they are not the same kind of thing at all. One can be tested from several directions. Does the cash flow statement agree with the profit line? Do the borrowings agree with the interest charge? Has working capital moved in the direction the revenue growth implies? A division publishes none of the statements those tests need, so not one of the tests is available on it. A multiple is applied to what the market can verify, and a segment line is dramatically less verifiable than a set of accounts.

Why does weak disclosure bite hardest on the division with the highest multiple?

Because of where the value sits relative to where the evidence sits, and the arithmetic here is worth doing slowly. Sankalp Industrial Systems Limited has three divisions. Division 1, industrial valves, turns over Rs 6,00,00,00,000 at a 23.0 per cent margin. Division 2, precision castings, turns over Rs 4,20,00,00,000 at 25.0 per cent. Division 3, aftermarket parts and service, turns over Rs 1,80,00,00,000 at 30.0 per cent. The margin gives Rs 54,00,00,000 of segment earnings before interest, tax, depreciation and amortisation (EBITDA).

All three figures are taken as given from the build covered separately in this sequence, along with the multiples applied to them: 7.5 times on division 1, 6.5 times on division 2 and 14.0 times on division 3. Choosing those multiples and defending them is covered separately; what matters here is what they do.

Division 3 across three different denominators shows the same business changing size dramatically depending on what it is measured against. Division 3 is 15.00 per cent of group revenue. The same division is 18.18 per cent of segment EBITDA. And it is 30.56 per cent of the gross parts value of Rs 24,73,50,00,000, before the head office line is taken off. The smallest division by revenue carries the largest share of the constructed value, which means the constructed value depends most heavily on the figures a reader can see least of.

That is not an accusation against anybody. The 30.0 per cent margin may be entirely real and the 14.0 times may be entirely defensible. The point is narrower and harder to escape: a market pricing Rs 54,00,00,000 that it cannot test from a balance sheet, cannot trace through a cash flow statement and cannot compare with a standalone audited peer does not pay 14.0 times for it with the same confidence that a fully reported standalone business would attract. Some of the discount is that difference in confidence, expressed as money.

Cause two: what does it cost when cash earned in one division can be spent in another?

This is the mechanism that surprises people, and the household example gets there faster than any diagram. The tuition money and the lodger's rent and the salary all land in one account. When the roof needs repairing, the money that pays for it comes out of that account. Nobody sits down and decides that the roof repair will be funded by tuition earnings specifically. The pooling happens first and the spending happens afterwards, and by then nobody can say whose money it was.

A group works exactly the same way at a much larger scale. The three divisions of Sankalp Industrial Systems Limited generate Rs 2,97,00,00,000 of segment EBITDA between them. Take off the Rs 9,00,00,000 of head office cost and the group reports Rs 2,88,00,00,000 of consolidated EBITDA. The consolidated figure is one pool. Who gets what out of that pool is decided inside the group, by people who are not required to explain the reasoning division by division.

So the aftermarket business can earn at a 30.0 per cent margin and see that cash spent installing a new line in the castings division that earns at 25.0 per cent. It may be an excellent decision. It may be a poor one. The point is that a holder outside the company can neither observe the decision as it is made nor prevent it, and that exposure is priced. Koller, Goedhart and Wessels make that argument when they connect what a business is worth to who decides where its capital goes, and it is the reason capital allocationDeciding which part of a business receives the cash the business generates. sits at the centre of the subject rather than at its edge.

Why is the discretion priced even in a year when no cash moves?

Here is the part that feels wrong on first encounter. Suppose the group has never once moved cash from one division to another. Every division has funded its own spending out of its own earnings for as long as anybody can remember. Does the capital allocation cause still bite?

The cause still bites, and the reason is that what gets priced is the discretion rather than the event. An outside holder is not exposed to what happened last year; they are exposed to what can happen in every year still to come. The absence of a transfer so far is evidence about the past, and weak evidence at that. A holder who cannot see the routing cannot confirm the absence either. Jensen and Meckling put the general version of this argument in place in 1976: when the people making a decision are not the people carrying its consequences, the gap between them has a cost, and that cost is real whether or not it is exercised in a given period.

A market prices what a holder is exposed to, and a holder inside a group is exposed to every future allocation decision, not only to the ones already made. The same logic runs through many other familiar places. A tenant on a month to month arrangement pays attention to the fact that the rent can be raised, even in a year when it is not raised. A supplier with one customer worries about the customer's freedom to switch, even while the orders keep arriving.

Cause three: what happens when nobody can buy the division they actually want?

Imagine a buyer of shares whose interest is confined to industrial services businesses. Call them Sthira Capital Partners, invented like everything else here. Sthira Capital Partners like exactly one thing about Sankalp Industrial Systems Limited: the aftermarket parts and service division, earning at a 30.0 per cent margin. Aftermarket service is the sort of business their whole approach is built around.

Sthira Capital Partners cannot have it. There is no route by which they can hold division 3 alone. To get the 30.56 per cent of the gross parts value that division 3 represents, they must take 100 per cent of the group. Taking the group means taking Rs 6,00,00,00,000 of valve revenue and Rs 4,20,00,00,000 of castings revenue they did not come for and do not want. For most buyers in that position the answer is simply to look elsewhere.

Now think about what that does to the price. Value is not set by the average opinion of everybody; it is set at the margin by whoever is willing to pay most. The holder for whom a business is worth most, given whatever else they hold, is its natural ownerThe holder for whom a business is worth most, given what else they hold.. A pure aftermarket services business has a set of natural owners who understand it, want more of it, and will pay accordingly. A bundle of one valve business, one castings business and one aftermarket business has to be wanted whole, so its set is much smaller. Thinning out the population of people who could plausibly want a thing moves the price, without changing the thing at all.

The same effect is familiar from selling second hand goods. A single dining chair has one narrow market. A set of six matching chairs has a different one. And a set of six chairs sold together with a wardrobe nobody asked for has a smaller market still. The buyer must want the whole lot or walk away. The wardrobe is not a bad wardrobe. It is simply attached to something else.

Cause four: why does a mixed group have no clean comparable?

The fourth cause is quieter than the other three and it lives inside the measurement rather than inside the business. Pricing anything by comparison requires something to compare it with. A pure valve manufacturer can be measured against Marudhar Valve Industries Limited. A pure castings business can be measured against Palani Castings Private Limited. An aftermarket services business can be measured against Bhima Flow Systems Limited, Chenab Industrial Products Limited or Tapti Service Partners Private Limited. All invented, and all single business companies whose whole reason for existing in this example is that they are one thing.

Against what does a mixed group get measured? Strictly, against another mixed group with the same three businesses in the same three proportions and the same head office arrangement. No such company is listed anywhere. Mahasagar Industrial Group Limited, also invented, may be a mixed industrial group, but its mix is not this mix, so comparing the two carries an error that has nothing to do with either company's quality.

The practical consequence is that a group's own blended multipleOne multiple describing a mixed group, here 8.34 times on the parts. is estimated more loosely than a single business multiple is. Sankalp Industrial Systems Limited trades at 7.78 times its Year 0 EBITDA of Rs 2,88,00,00,000, close to the 7.8 times median of the six invented peers in this sequence, and the parts imply 8.34 times on the same EBITDA. The difference is 0.57 turns computed on the unrounded figures; subtracting the two printed two decimal figures instead gives 0.56, and whoever states either number should say which arithmetic produced it. Neither multiple is wrong, and the width of the honest error band around a mixed group's own figure is itself part of why a gap appears.

The sharpest version of this whole argument is a holding companyAn entity whose main assets are stakes in other businesses., whose principal assets are stakes in other businesses rather than operations it runs itself. All four mechanisms above turn up there in a more extreme form, and the gap attached to one has its own name and its own treatment. Sankalp Industrial Systems Limited is not one of those: it is an operating group that reports three divisions, and that is the case worked throughout.

Is the Rs 9,00,00,000 of head office cost a measurement artefact?

No, and this is worth separating carefully from everything above, because it is the one line in the build that people are tempted to dismiss as bookkeeping. Segment EBITDA across the three divisions comes to Rs 2,97,00,00,000. Consolidated EBITDA is Rs 2,88,00,00,000. The Rs 9,00,00,000 in between is unallocated corporate costHead office cost belonging to no division, here Rs 9,00,00,000.: the group's own head office, belonging to no division and therefore carrying no multiple of its own.

The head office cost is entirely real. Somebody signs those salaries. The group secretary, the consolidation team, the listing obligations, the audit of the group as a whole: no accountant conjured any of that up to make a table balance. A household running three earning activities out of one home also has costs that belong to none of them, and the electricity bill does not become imaginary because it cannot be allocated fairly.

The choice is what to do with that cost in a parts build. The treatment used in the locked build capitalises the Rs 9,00,00,000 at the blended 7.8 times, giving Rs 70,20,00,000, and deducts it from the gross Rs 24,73,50,00,000 to reach Rs 24,03,30,00,000. A second, equally defensible treatment spreads the Rs 9,00,00,000 across the three divisions in proportion to revenue and carries each division's reduced EBITDA through its own multiple. Allocating by revenue sends Rs 4,50,00,000 to division 1 on its 50.0 per cent revenue share, Rs 3,15,00,000 to division 2 on 35.0 per cent, and Rs 1,35,00,000 to division 3 on 15.0 per cent.

Work it through: the second route lands at Rs 24,00,37,50,000, a full Rs 2,92,50,000 below the first, and turns the measured discount from 6.79 per cent into 6.68 per cent. Two defensible treatments of the same real cost give two different answers, and neither of them is more correct than the other. The alternative is computed here only to show that the two genuinely differ; which treatment to adopt is decided in the build covered separately.

How is a conglomerate discount actually measured?

By repetition, and that word is doing all the work. The measurement itself is trivial: build the parts, read the traded value, subtract, divide, name the base. Anybody can do it once in an afternoon. The result means something only when the same build is run again next period, with the same three multiples, the same treatment of head office cost and the same definition of the traded figure, and then again after that.

The reason is that a single figure cannot distinguish between at least three completely different situations. The gap may be persistent, sitting at roughly the same level period after period. A persistent gap suggests a structural feature that has not changed. The gap may instead be widening, and a widening gap is a different fact that invites a different question. Or the gap may have appeared because one of the three chosen multiples has gone stale while the market moved, in which case the measurement is reporting on the inputs and nothing else.

Only a series measured the same way each period can tell those three apart, and no snapshot ever can. Practitioners who take the exercise seriously therefore write down their three multiples and their cost treatment in a note, and then refuse to change them without recording why. If the method drifts, the series measures the analyst's changing mind rather than the market's changing price, and the whole point is lost.

Sankalp Industrial Systems Limited has exactly one measurement occasion behind it, the 6.79 per cent worked above, and one occasion is not a series. A made-up history of the gap would be exactly the fabrication the discipline is meant to prevent.

How much of this measured discount is an assumption?

Most of it, on this particular company, and the arithmetic is easy enough to check without a calculator. Division 3 has Rs 54,00,00,000 of segment EBITDA. Move its multiple by two turns, from 14.0 times to 12.0 times, and its value falls by two times Rs 54,00,00,000, or Rs 1,08,00,00,000. Nothing else in the build moves. The parts fall from Rs 24,03,30,00,000 to Rs 22,95,30,00,000, and the measured gap falls from Rs 1,63,30,00,000 to Rs 55,30,00,000. Off the new parts figure that gap is 2.41 per cent.

So two turns on the smallest division, applied to a figure of Rs 54,00,00,000 in a group turning over Rs 12,00,00,00,000, removed Rs 1,08,00,00,000 of measured discount. As a share of the original gap that is 66.14 per cent, or two thirds in round terms. Two thirds of this measured conglomerate discount came from one chosen multiple on one small division, and nothing about Sankalp Industrial Systems Limited was different at either setting.

Push the same lever a little further and the gap vanishes altogether. At about 10.98 times on division 3, precisely 10.9759 times, the parts land exactly on the traded Rs 22,40,00,00,000 and there is no discount to report. At 10.0 times the parts sit at Rs 21,87,30,00,000, or minus Rs 52,70,00,000 against the traded figure. The parts are now below the whole, and the same arithmetic that produced a discount produces a premium instead. The company is identical at every one of those settings. Only one input moved.

What is a conglomerate discount not?

A conglomerate discount is not a verdict. A conglomerate discount is a difference between two numbers, one of which was constructed out of four assumptions, and a difference is not the same object as a judgement. Everything above explains why groups get priced below the sum of their divisions, and the sentence every reader is waiting for still has to be refused.

So, plainly. A gap is not evidence that a market has got anything wrong. A market that prices a group below its divisions valued separately may be pricing exactly the four mechanisms described above, and pricing them sensibly. A gap is not an argument that the parts belong apart. Separating a division, selling one or restructuring a group is a transaction with a buyer on the other side, and a buyer's price is a different number built out of different evidence. And a number that moves this far on one chosen input cannot be read as an opportunity in either direction.

The equivalent refusal in a household is easy to feel. The tuition classes being hard to value separately does not mean the household should be broken up, and it does not mean somebody has undervalued the tuition classes. The difficulty means what it says: the classes sit inside something, and being inside something has consequences that can be named and, with care, sized. A group priced below its divisions valued separately is either priced for something the parts build does not capture or it is not, and a subtraction on its own cannot tell which.

One last note on precision. Every rupee figure above falls on a whole number of lakh by construction, so each one is exact rather than rounded. The percentages are not exact and are rounded once, from the full value, when printed: 6.79, 7.29, 2.41, 6.68, 66.14, 30.56. Rebuilding any of them from two already printed figures will sometimes land a digit away. The stray digit is a rounding artefact rather than a defect, and whoever rebuilds a percentage that way should say so.